Learning by Imitation using Inductive Logic Programming.

摘要

The transfer of knowledge from an expert to a software agent can be a tedious task. Instead, the agent can learn by observing the expert. This thesis examines, with RoboCup soccer data, how inductive logic programming can be used to learn rules in first-order logic that describe an agent's underlying behavior. Experimental results have shown that a discriminatory induction algorithm generates rules with better quality than those generated by a descriptive induction algorithm. We also show that the rules learned can be directly applied in a real-time environment with excellent response time. We conclude the thesis by comparing the results with previous CBR research work, and provide suggestions for future work.